[Colloquium] TTIC Colloquium: Ruslan Salakhutdinov, MIT

[Julia MacGlashan]

* Note: This talk is on *Tuesday* due to MLK Jr Day on Monday *


When:             *Tuesday, Jan 19 @ 1:00pm*

Where:           * TTI-C Conference Room #526*, 6045 S Kenwood Ave


Who:              * **Ruslan Salakhutdinov*, MIT


Title:          *      **Unsupervised Learning of Deep Probabilistic Models*



 Building intelligent systems that are capable of extracting high-level
representations from high-dimensional data lies at the core of solving many
AI related tasks, including object recognition, speech perception, and
language understanding. Theoretical and biological arguments strongly
suggest that building such systems requires models with deep architectures
that involve many layers of nonlinear processing.

In this talk I will first introduce a class of probabilistic generative
models called Deep Belief Networks that contain many layers of latent
variables. I will first provide a theoretical justification behind greedy
learning algorithm that learns one layer of latent variables at a time. I
will then show that these deep hierarchical models are able to learn useful
feature representations from large, unlabeled datasets. The learned
high-level representations can be used for subsequent problem-specific
tasks, such as object recognition, information retrieval, collaborative
filtering, or nonlinear dimensionality reduction.

In the second part of the talk I will describe a new hierarchical model
called a Deep Boltzmann Machine. Like Deep Belief Networks, Deep Boltzmann
Machines have the potential of learning internal representations that become
progressively complex at higher layers, which is a promising way of solving
object recognition problems. Unlike Deep Belief Networks and many existing
models with deep architectures, the approximate inference procedure, in
addition to a fast bottom-up pass, can incorporate top-down feedback. I will
further describe a new learning algorithm that combines variational methods
and MCMC and will demonstrate that Deep Boltzmann Machines learn good
generative models and perform well on handwritten digit and visual object
recognition tasks.
*
** Host:              Nati Srebro, nati at ttic.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20100112/8a670565/attachment.htm